Presentation Information
[P02-201]Establishing a 3D Structure- and AI-Based System for the Development of Novel Enzymes
○Satoshi Yuzawa1, Christopher J. Vavricka1 (1. Tokyo University of Agriculture and Technology (Japan))
Keywords:
CYTOCHROME P450,AI-BASED ENZYME EXPLORATION,REGIOSELECTIVE AROMATIC HYDROXYLATION,STRUCTURE-BASED ENZYME ENGINEERING
[Purpose]
Conventional methods to predict enzyme
function rely heavily on sequence homology
and existing annotations, limiting the ability to
discovery new enzymes for novel or poorly
characterized reactions.
To solve this problem, we developed GEnESIS
(Graph-based Enzyme Evolution with Structure-
Informed Scoring), an enzyme discovery
platform that integrates three-dimensional
structural information and AI to discover and
prioritize novel enzymes based on the structural
features that govern reaction selectivity. This
approach was then applied to identify
cytochrome P450 candidate enzymes with high
potential to catalyze the selective hydroxylation
of aromatic compounds.
[Method]
Substrates were docked into heme
proteins annotated as cytochrome P450 to
construct a protein-ligand complex library. Next,
a GNN was used to learn features representing
protein-substrate interactions. The selected
complexes were then redocked and refined into
reactive state models containing a highly
reactive intermediate. A reactivity prediction
model was subsequently built using structure-
derived information, and candidate rankings
were stabilized by integrating multiple models.
Finally, molecular dynamics simulations were
performed for the top candidates to evaluate the
stability of reaction-compatible spatial
arrangements.
[Results]
Using GEnESIS, promising enzyme
candidates could be ranked for regioselective
hydroxylation using multiple aromatic
substrates. Compared with conventional
methods, the candidates selected by GEnESIS
covered a broader sequence space and
enabled construction of a candidate library
while maintaining higher sequence diversity.
Furthermore, molecular dynamics simulations
confirmed that the top candidates selected by
GEnESIS maintained the spatial arrangements
suitable for reactivity. These results indicate
that, by incorporation of key structural
information, GEnESIS can effectively select
candidate enzymes suitable for the selective
hydroxylation of multiple aromatic substrates.
[Consideration]
These findings suggest that
inclusion of enzyme-substrate structural
information, in addition to sequence similarity, is
an important indicator for superior enzyme
functional prediction. Using a GNN to learn
features of protein-substrate interactions,
GEnESIS was shown to capture not only
binding properties but also the structural
requirements necessary for the actual reaction
to occur.
[Conclusion]
These results demonstrate that
GEnESIS is a method capable of selecting
promising enzyme candidates based on the
structural arrangement of the active site rather
than relying solely on sequence similarity.
Furthermore, GEnESIS was shown to be
applicable to the regioselective hydroxylation of
multiple aromatic substrates, suggesting that it
may be further extended to additional novel and
non-natural reactions.
Conventional methods to predict enzyme
function rely heavily on sequence homology
and existing annotations, limiting the ability to
discovery new enzymes for novel or poorly
characterized reactions.
To solve this problem, we developed GEnESIS
(Graph-based Enzyme Evolution with Structure-
Informed Scoring), an enzyme discovery
platform that integrates three-dimensional
structural information and AI to discover and
prioritize novel enzymes based on the structural
features that govern reaction selectivity. This
approach was then applied to identify
cytochrome P450 candidate enzymes with high
potential to catalyze the selective hydroxylation
of aromatic compounds.
[Method]
Substrates were docked into heme
proteins annotated as cytochrome P450 to
construct a protein-ligand complex library. Next,
a GNN was used to learn features representing
protein-substrate interactions. The selected
complexes were then redocked and refined into
reactive state models containing a highly
reactive intermediate. A reactivity prediction
model was subsequently built using structure-
derived information, and candidate rankings
were stabilized by integrating multiple models.
Finally, molecular dynamics simulations were
performed for the top candidates to evaluate the
stability of reaction-compatible spatial
arrangements.
[Results]
Using GEnESIS, promising enzyme
candidates could be ranked for regioselective
hydroxylation using multiple aromatic
substrates. Compared with conventional
methods, the candidates selected by GEnESIS
covered a broader sequence space and
enabled construction of a candidate library
while maintaining higher sequence diversity.
Furthermore, molecular dynamics simulations
confirmed that the top candidates selected by
GEnESIS maintained the spatial arrangements
suitable for reactivity. These results indicate
that, by incorporation of key structural
information, GEnESIS can effectively select
candidate enzymes suitable for the selective
hydroxylation of multiple aromatic substrates.
[Consideration]
These findings suggest that
inclusion of enzyme-substrate structural
information, in addition to sequence similarity, is
an important indicator for superior enzyme
functional prediction. Using a GNN to learn
features of protein-substrate interactions,
GEnESIS was shown to capture not only
binding properties but also the structural
requirements necessary for the actual reaction
to occur.
[Conclusion]
These results demonstrate that
GEnESIS is a method capable of selecting
promising enzyme candidates based on the
structural arrangement of the active site rather
than relying solely on sequence similarity.
Furthermore, GEnESIS was shown to be
applicable to the regioselective hydroxylation of
multiple aromatic substrates, suggesting that it
may be further extended to additional novel and
non-natural reactions.
Comment
To browse or post comments, you must log in.Log in